DRL-Based Energy Minimization in Fast HARQ With Finite Blocklength Codes and Feedback Delay

In this paper, a fast hybrid automatic repeat request (HARQ) protocol is employed, where some HARQ feedback is omitted and the associated channel uses are incorporated for data transmission. Building upon this strategy, point-to-point communication systems which emphasize low delay and high reliability are revisited. Based on the established findings in finite blocklength (FBL) codes, systems employing the fast HARQ protocol operation under low transmission delay constraints and under low queueing delay constraints are explored, respectively. Then, long-term bit energy minimization problems are formulated. In light of the non-convex nature of the problem and the presence of small decoding error probabilities, finite-episode Markov Decision Process (MDP) with double-layer penalty rewards are established. Subsequently, an actor-critic based deep reinforcement learning (DRL) algorithm is designed. Through numerical evaluations, it is shown that compared with the benchmark schemes, the proposed scheme is more energy efficient especially when the packet size is large. For the buffer-limited system with fast HARQ, there exists a coding rate that can optimize the energy efficiency.